03867ntm a22005417i 4500 000723258 CZ-PrVSE 20250621144446.0 m d cr n|||||||||| 250621s2025 xr fsbm 000 0 eng d NEZPRACOVANÝ IMPORT ABA006 cze ABA006 ABA006 rda Tcypin, Gleb ISIS:162153 dis Zkoumání faktorů ovlivňujících výsledky zápasů v Dota 2 eng Exploring predictors of results of Dota 2 matches / Gleb Tcypin 2025 ?? stran : digital, PDF soubor Vedoucí práce: Jan Rejthar Bakalářská práce (Bc.)—Vysoká škola ekonomická v Praze. Fakulta informatiky a statistiky, 2025 Obsahuje bibliografii Textový (vysokoškolská kvalifikační práce) Rok obhajoby 2025 This thesis investigates the prediction of professional Dota 2 match outcomes using machine learning techniques, focusing on pre-game and early-game features limited to the first 10 minutes of gameplay. Alongside the development of predictive models, it introduces a structured data governance framework inspired by modern data engineering practices, enabling scalable data collection, transformation, and modelling. Three classification algorithms are applied: Logistic Regression, Multi-Layer Perceptron (MLP), and XGBoost. Each model is trained on a dataset adjusted to suit its specific characteristics and working principles, incorporating custom transformations such as binning and one-hot encoding. The models undergo initial evaluation and are then refined through feature pruning and hyperparameter tuning. The impact of individual variables is assessed using model coefficients, permutation importance (measured on the training data), and Shapley values. The models achieve competitive accuracy, with all reaching up to 65.7%. Although pruning and tuning yield incremental improvements, none of the individual models surpasses the baseline Logistic Regression, which also reaches 65.7%. A stacked ensemble model that combines the best-performing versions of all three base learners slightly improves performance to 65.8%. Key predictors include early-game differences in experience, kills, and net worth, as well as the presence of a team on the ELO leaderboard. The results confirm that early-game data alone can support meaningful predictions and that combining ensemble techniques with structured preprocessing pipelines enhances both performance and interpretability. Future work may involve integrating real-time match data and team-specific historical information to further improve predictive accuracy. Způsob přístupu: Internet data analytics [obor bakal. práce] bakalářské práce fd132403 czenas bachelor's theses eczenas Machine Learning Data Governance XGBoost MLP Dota 2 Logistic Regression Rejthar, Jan ISIS:126443 ths Kavřík, Dominik ISIS:126085 opn Vysoká škola ekonomická v Praze. Fakulta informatiky a statistiky kn20010709399 dgg https://insis.vse.cz/zp/87944/podrobnosti VŠKP v InSIS https://insis.vse.cz/zp/87944 Hlavní práce https://insis.vse.cz/zp/87944/posudek/vedouci Hodnocení vedoucího https://insis.vse.cz/zp/87944/posudek/oponent/86656 Oponentura https://insis.vse.cz/zp/87944/priloha/31241 Přiloha k práci https://insis.vse.cz/zp/87944/podrobnosti dc:identifier NEPOSILAT VSKP vse87944 250618 87944